Radiomic analysis of high-resolution computed tomography predicts interstitial lung disease progression and mortality...

The automated assessment of chest high-resolution CT (HRCT) scans allows the objective quantification of interstitial lung disease (ILD)-related features. Here, we explored the ability of HRCT-derived radiomic features to predict all-cause mortality and ILD progression in SSc-ILD patients.

We analysed baseline and follow-up HRCT scans of SSc-ILD patients using lung texture analysis (LTA™, Imbio). Lung parenchymal alterations (honeycombing, ground-glass, reticulation, hyperlucency) and pulmonary vessel volume (PVV) were quantified as percentages of the respective lung volume-for whole lungs and for upper, middle and lower zones. Univariable and multivariable Cox regression and generalized estimating equation (GEE) models were applied to identify radiomics predictors of all-cause mortality and ILD progression, respectively.

Of 313 SSc-ILD patients, 267 (85%) were eligible for radiomic analysis and were included in the mortality analysis; 160 (60%) were included in the progression analysis. Over 42 [interquartile range (IQ)R 27-110] months of follow-up, 65 (24%) patients died. They presented with a greater extent of PVV% and parenchymal alterations across all lung zones, compared with patients who survived. PVV% independently predicted mortality [adjusted hazard ratio (HR) 1.21, 95% CI 1.094-1.354], particularly in the upper zones (adjusted HR 1.28, 95% CI 1.11-1.46). Among 261 yearly follow-up visits from 160 SSc-ILD patients, 50.6% showed at least one episode of ILD progression, for a total of 99 (37.7%) episodes. Baseline radiomic features were overall comparable between progressors and non-progressors. The extent of whole-lung honeycombing was identified as the only independent radiomic predictor of ILD progression [adjusted odds ratio 1.53, 95% CI 1.17-2.01].

The PVV% and the extent of honeycombing independently predict mortality and ILD progression, respectively. These radiomic features might support risk stratification of SSc-ILD patients, on top of functional and clinical characteristics.